Cleaned Data, Three Ways

• Plain text files in a common, open, non-proprietary file format are best practice
  • these kinds file formats enable exchangeability and give greatest flexibility for choice of tools/programs for downstream analysis
• There are many good tools for data analysis, including:
  • Excel is a general spreadsheet tool, with some statistical capability
    • Excel may need care when importing data, as it does not have statistical datatypes, and converts data to new forms silently.
  • Minitab is a general-purpose statistical package, with an intuitive graphical interface
  • R is a general-purpose programming language with an emphasis on statistics and data science
• Minitab and R have modes of operation that enforce good data practice; Excel does not
• R has an extremely wide range of libraries and packages for specialised and advanced statistical analysis, which are not available in either Minitab or Excel

1 Setup

To participate in this workshop or to work through the example in your own time, you will need the following on your own computer:

1. The cleaned OpenRefine dataset from the “Cleaning Data With OpenRefine” workshop: esoph-tab.tsv (or whatever you named it)
2. Microsoft Excel (you have a licence for this via the university)
3. Minitab (you have a licence for this via the university)
4. R and RStudio

Please see the configuration notebook for help and guidance on setting up R and RStudio.

To download and install the latest versions of Excel and Minitab please visit the university’s IT pages.

2 Introduction

Data can be collected, stored, and analysed in a variety of ways, using a wide range of tools. We do not prescribe any particular approach, other than to note best practices for FAIR and op science prefer open, plain-text, human-readable non-proprietary data formats that are shared in public repositories. The best tool for data analysis for your work may differ depending on your area, and from project to project. Most good tools will allow you to read and write data in exchangeable formats compatible with FAIR and open science. In this workshop we will explore three common tools: Excel, Minitab and R, and use these to open and analyse our cleaned dataset esoph-tab.tsv from the earlier workshop.

We will perform three main tasks in each package:

1. Loading (or importing) the dataset
2. Generating standard summary statistics: mean, standard deviation, quartiles, etc.
3. Visualising the distribution of the data graphically

3 Excel

Microsoft Excel is almost ubiquitous. It is now available free of charge as a cloud service, as part of Office365¹. Alternatives to Excel, such as Google Sheets² and Libre Office³, work in much the same way. This workshop should be adaptable to those alternatives with minimal modification.

3.1 Loading data

1. Start Excel
2. Click on Open
3. Navigate to the location of esoph-tab.tsv, select it and click Open

For any format other than its own, proprietary .xls or .xlsx format, Excel will start the Text Import Wizard.

This makes loading the simplest datasets seem more complicated than it really is, and serves to nudge users towards Microsoft’s own format, and away from good data practice.

The Text Import Wizard should correctly identify that your data are delimited⁴.

However, Excel may not notice that the first row is a header row⁵, and may treat all rows equivalently.

1. Click on Next >

On the next page of the wizard, Excel indicates that it has identified the tab (\t) character as the column delimiter. You would be able to select a different character if this was incorrect. Here, no changes are necessary.

1. Click on Next >

In the next wizard page, Excel allows you to select each column’s data format. These formats are: General, Text, Date, and “skip column”.

Excel does not allow you here to specify data formats or data types useful for statistical or data analysis.

In particular, it does not distinguish between categorical and numerical data. It does not allow you to specify whether numerical values should be counts (whole numbers) or “real” values (can take any decimal value).

Importing data into Excel has very limited data validation.

1. Click Finish

Excel now presents your data in spreadsheet format.

Here, Excel again attempts to nudge us towards using one of the proprietary Excel formats by claiming that there is **Possible Data Loss:** Some features might be lost if you save this workbook in the comma-delimited (.csv) format. To preserve these features, save it in Excel format.

This is misleading. Our data should “safe” so long as we keep it in plain text, open, non-proprietary formats.

1. Would it be better if Excel were able to distinguish header rows from data rows?
2. What kinds of data do we have in each column? (numerical or categorical? count or real-valued?)
3. Did Excel import our data correctly?

Excel did not import our data correctly.

Excel interpreted the range 10-19 as a date: October 19th. Our tobgp column now has a number of dates interspersed among the values. This is bad: Excel has changed our data without asking or notifying us that it would do so.

To avoid this problem, we must specify Text as the data format for our columns:

1. Close the spreadsheet.
2. Start Excel
3. Click on Open
4. Navigate to the location of esoph-tab.tsv, select it and click Open
5. Click on Next >
6. Click on Next >
7. Click on each column in turn and select Text

1. Click on Finish

Excel has now imported our data, and it looks to be correct.

1. Did Excel import our data correctly?

3.2 Summary Statistics

To obtain summary statistics for each column we need to use Excel’s functions and pivot tables, as we need to treat categorical data differently from numerical data. We want to know, for instance, the number of datapoints we have in each agegp category, rather than a “mean” of the categories (which has no real meaning); but for our cases and controls, we might still want to know means, standard deviations, etc.

In all cases, we are potentially modifying our cleaned dataset directly because we are making changes to the same file we loaded - this would be bad practice.

Our first step should be to save a new spreadsheet to do our analysis in

1. Save the current file as an Excel workbook (note: this is necessary to avoid losing Excel’s calculations and formatting)

3.2.1 Summary statistics for quantitative/numerical data

We will calculate mean and standard deviation values for columns ncases and ncontrols. To do so, we need to use Excel’s AVERAGE and STDEV.S functions.

• The ncases values run from cell D2 to cell D89
• The ncontrols values run from cell E2 to cell E89

1. In cell D90 enter the text =AVERAGE(D2:D89) and hit the return key. This will calculate the mean of quantitative data in the column of cells extending from D2 to D89, and populate

1. Did Excel calculate the average value? If so, what is it?

Excel did not calculate the mean value.

When we selected Text as the input data format, Excel used this to do two things:

1. All numerical values are now understood as text, not as numbers⁶.
2. All cells in both columns - all the way to the bottom of the spreadsheet - are interpreted as text, not numbers, formulas, or any other element.

To rectify this problem, we need to change those two columns to General format:

1. Select all numerical data in columns D and E (D2:E89)
2. Click on the warning icon (a small yellow triangle with an exclamation mark)
3. Click on Convert to Number

1. Does the formula now give a plausible mean value?
2. If you were new to this file, how could you tell that the formula gives a mean value, and not a standard deviation, or something else entirely?

1. As there is no visual indication in the file of what the number in cell D90 represents, add a label to cell C90 with the text Mean:.
2. Copy and paste the formula from cell D90 into cell E90
3. Add the label Std Dev: to cell C91
4. Enter the formula =STDEV(D2:D89) into cell D91
5. Copy and past the formula from cell D91 into cell E91

Now we have some summary information for our quantitative data.

3.2.2 Summary statistics for categorical data

It is often useful to know how many datapoints you have in each category. Here, we could independently sum each of the ncases or ncontrols cells, depending on which mixture of categorical variables we wanted (e.g. combine all cells for rows where tobgp is equal to 10-19), but Excel provides a tool called a pivot table to make things easier for us.

For instance, suppose we wanted to know the numbers of cases and controls in each category of the alcgp column. We could insert a pivot table that combined all values from ncases and ncontrols, for each of the individual categories in alcgp. To do this:

1. Select all of your data (A2:E89)
2. Click Insert \(\rightarrow\) Pivot Table \(\rightarrow\) OK

This brings up an interface that might at first look confusing. We are going to try to create a table that has one row per category in alcgp, and holds three columns:

• The alcgp category
• The sum of values in ncases for that category
• The sum of values in ncontrols for that category

To do so:

1. Click on the checkbox next to alcgp - this will place it in the Rows box, and you will see a view of the new pivot table
2. Click on the checkbox next to ncases - this will place it in the Values box as Sum of ncases and you will see the pivot table change
3. Click on the checkbox next to ncontrols - this will place it in the Values box as Sum of ncontrols and you will see the pivot table change

1. How many cases are there in total?
2. How many controls are there in the 120+ category
3. Why do you think the column of category names is not sorted in numerical order?

3.3 Data visualiation

Suppose we want to compare the numbers of cases and controls for each of our alcgp categories, we can reasonably use a bar graph (though we should really use a 1D scatterplot). To do so:

1. Select the data in the pivot table (A3:C7 in the figure)
2. Click on Insert \(\rightarrow\) Pivot Chart

This will create a bar chart in the current worksheet, showing the number of controls and cases in each category of alcgp.

4 Minitab

Minitab is a statistical software package, often used for teaching. It is moving towards a cloud-hosted model, but the Minitab Express package is still available for desktop use. This software requires a licence, which you will need to obtain from the university before you can run the tool.

Also, because of licensing restrictions, you must be connected to the university VPN so that the licensing server can be reached. Otherwise, you will not be able to use Minitab.

4.1 Loading data

1. Start Minitab Express
2. Click on File \(\rightarrow\) Open Worksheet
3. Navigate to the location of esoph-tab.csv, select it and click Open

Minitab Express will not import tab-separated value files.

Minitab Express presents your data in a panelled window. By default, the data is presented in spreadsheet form in the lower panel. The top panel will contain output, and the left panel will be a Navigator. This layout reflects Minitab’s role as a statistical analysis package, rather than a general spreadsheet tool.

Minitab Express did not ask any questions about data type or formats.

To see how Minitab Express imported your data:

1. Click on any column in the spreadsheet view
2. Click on the Inspector button (top right)

A new window appears. Click on the Column icon (third icon).

1. Has Minitab Express correctly distinguished between the header and data rows?
2. What data type or format does Minitab Express think each column is?
3. What kinds of data does Minitab Express “know about”?
4. Do you need to change any of the data values or formats?
5. How does this compare to loading data into Excel?

4.2 Summary Statistics

As a statistical package, Minitab Express is written to make obtaining summary statistical data relatively straightforward. Here we will try to recover the same data as we did in Excel.

4.2.1 Summary statistics for quantitative/numerical data

To obtain summary statistics for a single data column in Minitab Express:

1. Click on Summary \(\rightarrow\) Column Statistics
2. Select C4 ncases and click on the arrow next to Variable
3. Click OK

Minitab then updates the spreadsheet with calculated values and presents some output in the top panel, describing what values were calculated, and which cells in the spreadsheet hold output values.

To generate a set of descriptive statistics for a data column:

1. Click on Summary \(\rightarrow\) Descriptive Statistics
2. Select C5 ncontrols and click on the arrow next to Variable
3. Click OK

Minitab doesn’t put these values in the spreadsheet, but it does show a nicely-formatted table with a number of descriptive statistics. You can switch between views on the data by clicking on previous steps in the Navigator (on the left, by default).

1. Does Minitab give you the same information/values as Excel for this data?
2. Does Minitab give you any useful extra data?

4.2.2 Summary statistics for qualitative data

We will again try to obtain case and control counts, conditioned on the alcgp category.

1. Click on Graphs \(\rightarrow\) Bar Chart
2. Select Mean or other function of a continuous variable
3. Click on Multiple Y variables \(\rightarrow\) Clustered
4. Choose ncases and ncontrols as the continuous variables
5. Choose alcgp as the categorical variable
6. Select Sum as the function

Minitab, unlike Excel, shows both the tabular summary and the bar graph visualisation in the upper panel.

1. Was the data import and summary process easier and more convenient in Excel or Minitab? Why do you think that was?

5 R

R is a general programming language, designed to be used by statisticians and data scientists. Unlike Excel and Minitab it is free and open-source. RStudio is a free, open-source integrated development environment (IDE) designed for use with R. It has powerful version control and project management features, and a very large number of third-party libraries are available for statistical, bioinformatics, and other activities. It is the de facto language of choice for much statistical and bioinformatics work in academia.

5.1 Loading data

1. Start RStudio
2. In the Files panel, navigate to the location of the esoph-tab.tsv file
3. Click on the esoph-tab.tsv file

Clicking on the esoph-tab.tsv file shows the contents of the file in a navigation window (top left by default), but does not load or import the dataset

Working in R/RStudio is unlike working in a point-and-click environment such as Excel or Minitab. Commands are issued by typing them in the Console window (bottom left, by default) or by writing and running scripts.

Today, we will write and run a simple script

Before we load the data, we need to make sure we are in the correct working directory, and create an empty script.

1. In the Files panel, click More
2. Click Set as Working Directory

In the Console window, you should see that a command has been run.

---

1. https://office.live.com/start/Excel.aspx↩︎

2. Google Sheets is a cloud-based spreadsheet package, free for personal use↩︎

3. Libre Office is a free, open office quite intended as a drop-in, free replacement for Microsoft Office↩︎

4. delimited: separated by a character that indicates a new column is starting. This is often a tab (\t) or comma (,)↩︎

5. header row: a row in a table that indicates the contents of each column↩︎

6. In Excel, numbers are right-aligned, text is left-aligned. Each cell also has a green triangle in its upper left corner to indicate that it is a text value, not numerical↩︎